Emotion in Human-Agent Interfaces

Recent years have seen an upsurge of interest in the implications and applications of emotional behaviour in the human-computer interface (Picard 1997). In particular, research has focused on interactive systems involving visual and/or vocal modes of communication, where the user and/or the system may exhibit both positive and negative expressive behaviour in the course of an ongoing dialogue. Such behaviour can lend naturalness to an otherwise artificial situation, but it can also be exploited to enhance the communicative effectiveness of a human-agent interface; for example, a system may detect the emotional state of a user (from the look on their face, or from the tone of their voice) and act accordingly, or it may signal a problem by adopting suitable emotional behaviour itself (by generating appropriate facial and/or vocal expressions). Current research in this area revolves around virtual manifestations of human-agent interfaces known as ‘Embodied Conversational Agents’ (ECAs): speech-enabled animated characters that can act for example as a receptionist, a tour guide or a personal tutor (Cassell et al 2000, Pelachaud and Poggi 2001, Beskow et al 2005). The ability of an ECA to detect and exhibit emotional behaviour is believed to be an important requirement for achieving effective and naturalistic human-computer interaction (Bates 1994).